What distinguishes boosts from nudges?

Many policy interventions can be used to change behaviour, including mandates or bans (restricting or eliminating options), fiscal measures (monetary incentives and disincentives) or non-regulatory and non-fiscal measures. In recent years, the latter category has increasingly been equated with nudging. Nudges come in different forms. They can be, in the terms of Sunstein (Reference Sunstein2016), ‘educative nudges’ (e.g. disclosure requirements, warnings, labels or reminders) or ‘non-educative nudges’ (e.g. default rules, ordering of items on a menu or website or cafeteria design). Non-educative nudges can target or enlist cognitive or motivational biases (e.g. inertia, procrastination and loss aversion; see Rebonato, Reference Rebonato2012) to change behaviour.

Recently, Grüne-Yanoff and Hertwig (Reference Grüne-Yanoff and Hertwig2016; see also Hertwig & Grüne-Yanoff, Reference Hertwig and Grüne-Yanoffin press) proposed that a distinction be drawn, on conceptual grounds, between boosts and nudges. They defined boosts as interventions that

 Rather than merely presenting pertinent and accurate information (as educative nudges do), boosts explicitly seek to foster existing decision-making competences and to develop new ones, thus enabling individuals to translate their intentions (preferences) into behaviour – that is, to exercise personal agency. Clearly, information can be instrumental in developing an intention to engage in a specific behaviour. Such intentions are not necessarily translated into action, however. Findings on the ‘intention–behaviour gap’ suggest that conceptual knowledge often does not suffice, but that it needs to be supplemented with, for instance, procedural knowledge (including self-regulatory strategies; e.g. see Abraham et al., Reference Abraham, Sheeran and Johnston1998).

Boosts thus differ from nudges in terms of the underlying assumptions about the potential value of educative efforts, the emphasis placed on the importance of actually exercising the power of choice (as opposed to being afforded the opportunity to choose), the levers used in the intervention (Hertwig & Grüne-Yanoff, Reference Hertwig and Grüne-Yanoffin press) and the underlying understanding of human decision-making (Grüne-Yanoff & Hertwig, Reference Grüne-Yanoff and Hertwig2016). Furthermore, although proponents of non-fiscal and non-regulatory approaches often share a commitment to Herbert Simon's notion of bounded rationality, their interpretations of its implications differ. Those in favour of nudges typically invoke the findings of Kahneman and Tversky's heuristics and biases research programme (see Rebonato, Reference Rebonato2012), emphasising decision-makers’ systematic cognitive biases and motivational shortcomings. Those in favour of boosts typically invoke findings showing that bounds on people's time, knowledge and computational powers do not prevent them from making good decisions, to the extent that they succeed in employing simple decision strategies in the appropriate contexts – that is, where there is a fit between cognition and environment (i.e. ecological rationality; Gigerenzer et al., Reference Gigerenzer, Hertwig and Pachur2011).Footnote ¹

For the reasons outlined above, this article argues that it is both useful and important to understand boosts as a distinct category of behaviour change tools. The goal is not, however, to engage in a terminological debate about nudges and boosts (for a conceptual analysis, see Grüne-Yanoff & Hertwig, Reference Grüne-Yanoff and Hertwig2016). Instead, this article aims to identify rules that policy-makers could apply in order to determine the conditions under which boosts, as defined above, are preferable to nudges.

Welfare, cost–benefit analyses and autonomy

How, at least in theory, do policy-makers choose which intervention to select? From the standpoint of nudging, the key criterion is welfare. According to Sunstein's (Reference Sunstein2014) broad definition, the term ‘welfare’ refers to “whatever choosers think would make their lives go well” (p. 73). Indeed, in many important situations (e.g. choices between medical treatment options or savings levels), what is primarily at stake is the choosers’ welfare. Their choices do not affect others, except perhaps indirectly. One way to approach the welfare question is to analyse relevant costs and benefits from the standpoint of the chooser and the policy-maker, and to select, among the possible interventions, the one maximising the chooser's welfare.

Two components of that analysis are the costs of decisions and the costs of errors for the chooser and the policy-maker. For example, a paradigmatic nudge such as a default rule (e.g. automatic enrolment in a pension plan) typically reduces the cognitive and other costs of decisions for choosers, who are freed from devoting time and attention to the problem. Similarly, as long as the rule is not especially difficult to devise, it also reduces costs for policy-makers, whose actions can have a broad and enduring impact in exchange for a relatively small investment (e.g. changing a legal default; see Chetty et al., Reference Chetty, Friedman, Leth-Petersen, Nielsen and Olsen2014).

Consider, for illustration, the goal of promoting healthy eating. For the policy-maker, it would be costly and time consuming to teach children and adults nationwide how to interpret nutrition labels, or to equip them with the numerical processing skills that are relevant for food choice (e.g. portion-size estimation skills; Dallacker et al., Reference Dallacker, Hertwig, Peters and Mata2016). It is likely to be less costly and more effective to change, where possible, the choice architectures in which food choices are made (Wansink, Reference Wansink2014). School cafeterias, for instance, could be redesigned to present the options differently. Salad bars could be moved away from the wall and placed in front of the cash registers, thus nudging students to make healthy choices. This and other relatively inexpensive changes to the choice architecture of all school cafeterias (e.g. what is displayed at eye level) could reach a large population of choosers.

Although setting up default rules is relatively inexpensive, default rules may also increase the costs of errors (understood in terms of the number and magnitude of mistakes) if the policy-makers who devise them only know what fits the statistically average person, but not what fits the individual. An organ donation default (Johnson & Goldstein, Reference Johnson and Goldstein2003), for instance, automatically makes everybody a potential donor, including those who, for whatever reason, do not wish to be donors. If the latter fail to opt out of the organ donation default, perhaps because of inertia or perceived social pressure, the default will lead to errors. (I bracket the fact that the rules for organ donation affect third parties.)

Boosts can, but need not, impose high costs for choosers and policy-makers. They can be very low in cognitive costs for the chooser – for instance, when transparent risk communication is used to foster the understanding of health statistics (e.g. Hoffrage et al., Reference Hoffrage, Lindsey, Hertwig and Gigerenzer2000; Gigerenzer et al., Reference Gigerenzer, Gaissmaier, Kurz-Milcke, Schwartz and Woloshin2007) or when people are taught to “strategically call on automatic processes” in order to translate self-declared goals into simple action plans (i.e. harnessing implementation intentions; Gollwitzer, Reference Gollwitzer1999, p. 493). Alternatively, boosts may require some hours of instruction and practice – for instance, training financial decision-making skills by teaching people simple financial and accounting heuristics (Drexler et al., Reference Drexler, Fischer and Schoar2014) or implementing a 12-hour sexual assault resistance training programme (including risk assessment strategies) for first-year female university students (Senn et al., Reference Senn, Eliasziw, Barata, Thurston, Newby-Clark and Hobden2015). To the extent that only those people who seek the competence offered by a boost will adopt it, this approach can be expected to reduce (and perhaps even eliminate) the costs of errors.

The analysis of costs and benefits also requires attention to the effectiveness of the intervention, usually measured in terms of the magnitude of its impact. Take, for instance, the objective of increasing people's retirement savings. Savings behaviour could be increased by efforts to promote financial literacy (boosting) or by default rules such as automatic enrolment (nudging). There is an ongoing debate about the effectiveness of these two distinct strategies. Willis (Reference Willis2011) has argued that financial literacy education has, at best, limited success. Should efforts to promote financial literacy indeed fail to meaningfully promote retirement savings, the argument against financial literacy boosts, and in favour of defaults, will be strengthened.Footnote ²

For the welfarist, integrating factors such as the costs of decisions, the costs of errors and the effectiveness of the intervention seems, at least in theory, straightforward: which approach has the highest net benefits? Policy-makers could quantify the costs and benefits of (say) default rules and compare them with the corresponding costs and benefits of boosting people's competences. From the perspective of political philosophy, of course, many people are not welfarists. Instead, they emphasise the importance of autonomy. Paradigmatic nudges such as defaults and boosts preserve autonomy in the formal sense; they do not prevent people from going their own way. But if autonomy is an important goal and, in particular, if the policy-maker's aim is to promote individual agency, there is an argument in favour of boosts. This argument is further strengthened by the possibility of errors caused, for instance, by defaults. For those who emphasise autonomy, the production of errors by passive rather than active decisions may represent a problem that cannot be offset by considerations of cost–benefit efficiency. More generally, whenever actual choice and, in particular, informed choice – rather than the formal opportunity to choose – are believed to be important because they are conducive to welfare (e.g. in terms of procedural satisfaction; Frey et al., Reference Frey, Hertwig and Herzog2014) or because they are intrinsic goods, boosts should be favoured.

Rules for policy-makers

Next, I will outline some additional, less-discussed criteria that could be invoked in the process of selecting an intervention. My focus will be on the choice between boost and nudge interventions. In principle, however, the following rules could be extended to include other kinds of interventions as well. I will list six rules. The first four relate to what could be interpreted as necessary requirements for boosting and nudging to succeed, measured in terms of welfare. If these requirements are not met, this shortcoming cannot be offset by any other advantage an intervention may have. The remaining two criteria could and should be included in the cost–benefit calculus outlined above and thus balanced against an intervention's other costs and benefits.

In order to develop new skills and competences, people need sufficient levels of motivation and cognitive capacity (Grüne-Yanoff & Hertwig, Reference Grüne-Yanoff and Hertwig2016). Consider, for example, the ability to make Bayesian inferences, which is often required in the context of medical choices. Assume that a patient's routine PSA screening test has produced a positive result. To understand the meaning of that test result, several pieces of information are needed: (i) the base rate of prostate cancer, p(PC); (ii) the probability that the test is positive if the person has prostate cancer (sensitivity or true-positive rate of the test; p[pos|PC]); and (iii) the probability that the test is positive if the person does not have prostate cancer (false-positive rate or 1 – specificity; p[pos|no PC]). On the basis of this information, the patient and his physician can use Bayes’ rule to calculate the positive predictive value (PPV) of a test; that is, the probability of someone who tests positive actually having prostate cancer:

For instance, if the base rate is about 6.3% and the test has a sensitivity of about 21% and a false-positive rate of 6%, the PPV is about 19%, which means that 81% of positive test results are false-positives (these numbers are realistic; see Arkes & Gaissmaier, Reference Arkes and Gaissmaier2012). Many doctors and patients do not understand these calculations. Consequently, doctors often cannot advise their patients properly (Gigerenzer et al., Reference Gigerenzer, Gaissmaier, Kurz-Milcke, Schwartz and Woloshin2007). One solution is to boost doctors’ and laypeople's statistical prowess by training them to translate probabilities (or percentages) into a representation that makes it much easier to calculate the PPV.

In the present example, that would mean replacing conditional probabilities by natural frequenciesFootnote ³ as follows:

Sedlmeier and Gigerenzer (Reference Sedlmeier and Gigerenzer2001) trained people to construct such frequency representations and compared the success of this approach with that of traditional and explicit rule training (i.e. teaching people to insert probabilities into Bayes’ rule, as is typically done in schools). Transfer of learning to new problems was good in both rule training and representation training. However, the latter showed greater temporal stability, with no decrease in performance even after 15 weeks. This kind of training in statistical and risk literacy is relatively simple and requires a one-off time investment of less than 2 hours. However, it does necessitate some basic cognitive abilities (e.g. numerical and arithmetic skills), as well as the motivation to learn to translate information from one representation into another. Without those, this statistical reasoning boost will not work.

Motivation and basic cognitive skills are also the prerequisites for another simple boost: a 5–10-minute expressive writing exercise has been shown to reduce test anxiety (Ramirez & Beilock, Reference Ramirez and Beilock2011), a widespread phenomenon (80% of community college students report a moderate to high degree of mathematics anxiety) related to negative outcomes such as impaired financial planning. The motivational demands for boosts that require a greater time investment will, of course, be even greater. In a study using a simple app, parents who were habitually anxious about mathematics did bedtime mathematics with their children, reading short numerical story problems and answering questions on topics such as counting, geometry, arithmetic, fractions and probability over the course of the school year. By the end of the year, these children outperformed children in a control group by almost 3 months in mathematics achievement (Berkowitz et al., Reference Berkowitz, Schaeffer, Maloney, Peterson, Gregor, Levine and Beilock2015). This is a great improvement, but it is conditioned on regular investment of time.

To conclude, if individuals’ cognitive resources are, for whatever reason, severely compromised or their motivation is low, boosts that require one or both of these resources will fail. In some contexts, policy-makers may instead be able to choose a boost that requires minimal cognitive competences (e.g. Gollwitzer, Reference Gollwitzer1999). But if even such competences cannot be assumed, then other kinds of interventions, such as default rules, appear preferable.

Such problematic cognitive and motivational conditions, however, also pose a challenge for policy-makers who decide in favour of nudging. Nudging (or soft paternalism; Sunstein, Reference Sunstein2014) is suggested to be less intrusive than mandates and bans because it imposes low costs and permits people to easily reverse the choice, thus maintaining freedom of choice. Reversibility is easy in theory but may prove difficult in practice, particularly if cognitive and motivational resources are compromised or unavailable. Under such circumstances, advocates of default rules must face the possibility that they will stick, even if they are ill suited to people's preferences.

In imposing a default rule, policy-makers (choice architects) seek to steer the chooser's behaviour towards his or her ultimate goal, as judged by him or herself (e.g. greater savings or healthier food choices). This requires the policy-maker to have information about the chooser's goals; otherwise, the default may not fit those goals, leading to errors. But policy-makers do not necessarily know what individuals care about (Rebonato, Reference Rebonato2012). Experts may even systematically misconstrue what people want for themselves – for instance, imposing a treatment-at-all-costs medical model that ignores quality-of-life concerns (Gawande, Reference Gawande2014). Even choosers themselves may not always be aware of their goals; sometimes they may need to work them out, and to do that, they need transparent information and the competence to process it. Let us again take participation in cancer screening tests as an example (e.g. mammography or PSA). Depending on the quality of the test (i.e. sensitivity and specificity) and the subsequent treatment options, an individual may decide to participate in a specific screening test (e.g. screening for colorectal cancer), but not another (e.g. screening for prostate cancer). In light of this test-dependent preference to participate, nudging people to participate in all cancer screenings – for instance, by using gain-framed (survival) rather than loss-framed (mortality) messages – would be ethically controversial. There is a good chance, moreover, that it would not increase people's welfare (see Arkes & Gaissmaier, Reference Arkes and Gaissmaier2012). Here, boosting – in terms of fostering statistical literacy – would be more appropriate (assuming the boost is effective).

A related problem arises when goals are heterogeneous across a population or in conflict within a person. One-size-fits-all default rules that do not fit an individual's conflicting goals (e.g. saving for retirement versus paying for a child's education) or people's diverse situations, concerns or values may nevertheless ‘stick’. Personalised default rules can reduce the risks (Sunstein, Reference Sunstein2014), but choice architects need a great deal of information to design such rules, and this information may well be lacking.

To conclude, when goals are uncertain, heterogeneous within a person or across a population or are even in conflict within a person, boosting people's competence (e.g. risk literacy or decision skills) is likely to be more appropriate than nudging, because the boost option is less prone to error.

Regulatory and fiscal interventions (e.g. a law prohibiting people from riding motorcycles without helmets or the taxing of cigarettes or soft drinks) are highly visible. Visibility is an important safeguard against arbitrary and unreasonable government interventions: citizens can scrutinise them and hold the government accountable for them. Nobody is confused or fooled. Boosts are visible because they require the audience's engagement. Many nudges are also visible and public (e.g. graphic health warnings or automatic enrolment in savings plans).

But some nudges, such as the order of items on a menu, may ‘fly under the radar’ of the person being nudged, potentially influencing their behaviour without their awareness or the recognition that they could act differently. For example, people appear not to realise that the amount they eat is substantially driven by the portion size of a snack. As a result, they do not adjust their subsequent intake to compensate for consuming a larger amount (Wansink, Reference Wansink2014). Similarly, shoppers may not be aware that they pay more attention to, and are much more likely to buy, products placed at eye level than products placed above or below it (‘eye level is buy level’). That is, the mechanisms that retailers use to ‘nudge’ shoppers may be invisible to consumers.

If the desired behavioural effects of a public policy nudge, such as the order of items on a choice menu, depend on people not being told that this intervention had been implemented – because, if informed, they would perceive it to be manipulative and categorically override it – then a boost has significant advantages, at least on ethical grounds, and may for that reason be more appropriate. How people respond when being told about either an intervention or its underlying mechanisms is an empirical question (for evidence, see Loewenstein et al., Reference Loewenstein, Bryce, Hagmann and Rajpal2015), but also raises ethical concerns.

Some have argued that it is not ethically necessary for governments to specify that an intervention has been implemented, especially if it would make the intervention less effective (see Bovens in House of Lords Science and Technology Select Committee, 2011). From this standpoint, it suffices that a perceptive person could discern for him or herself that an intervention is in place. However, this criterion may be too weak for several reasons. First, a person's ability to discern an intervention as such (e.g. a default or a different order) is distinct from the ability to discern how it changes his or her behaviour – particularly if the direction of the effect is counterintuitive. Second, if most people are, in practice, unable to discern how an intervention changes their behaviour, its underlying mechanism will essentially be hidden, rendering it difficult for people to reverse their choice and compromising their freedom of choice. Third, if most people who are told about a nudge and its targeted effects would reject it, then the nudge lacks acceptability and public authorisation.

More generally, if an intervention or its underlying mechanism is invisible, it does not treat people with respect, and it may undermine their dignity and agency. In a free society, I believe, public officials should generally follow a version of John Rawls’ publicity principle: they should not adopt policies that they could not defend in public. If a transparent intervention ran into public opprobrium, there would be good reason not to adopt it. Furthermore, in a free society, it should ideally be difficult to keep things hidden, and an intervention that officials try to hide will hopefully come into the open – compromising the entire enterprise.

Choice architectures are ubiquitous and their systematic impacts can rarely be avoided. The private sector nudges, as does the government. There is little doubt that both private and public choice architectures can have illicit goals or be manipulative (e.g. Nestle, Reference Nestle2015). Governments influenced by wealthy donors or lobbyists, for example, may act in a way that does not have the welfare of the citizens at heart, but that favours particular interest groups. Thus, nudges used by public officials may, in the worst-case scenario, be coercive and manipulative (Rebonato, Reference Rebonato2012). To promote both welfare and autonomy under such circumstances, it is desirable for individuals to be equipped with the competence to see through and to counteract illicit or manipulative nudges – and toxic choice environments more generally.

Take, for instance, the problem of asymmetric framing in the medical domain. One widely used industry technique is to report the benefits of medical interventions in the form of relative risks (big numbers) and their harms and side effects in the form of absolute risks (small numbers). This asymmetry magnifies the representation of benefits and minimises harms. To inoculate patients against this industry nudge, they should be equipped with the skills to see through this and other manipulative techniques and to understand both absolute and relative risks and health statistics more generally (Gigerenzer et al., Reference Gigerenzer, Gaissmaier, Kurz-Milcke, Schwartz and Woloshin2007).

The phenomenon of defensive decision-making is another example of how a choice environment can become toxic. Although highly prevalent in medicine, defensive decision-making is not limited to this field. In medicine, it describes the behaviour of doctors who alter their clinical decisions in response to the threat of being sued for medical malpractice. The option they pursue is not necessarily the best for the patient, but is motivated by the threat of liability. Defensive practices include ordering more diagnostic tests, prescribing more medication than is medically indicated or suggesting unwarranted invasive procedures. Such practices are prevalent in high-liability specialties (e.g. gynaecology, radiology and orthopaedic surgery) in the USA (Studdert et al., Reference Studdert, Mello, Sage, DesRoches, Peugh, Zapert and Brennan2005), but are also relatively common in less litigious health environments. Within such an environment, patients should not blindly follow their doctors’ advice; rather, they also need to understand the relevant numbers (e.g. health statistics).

To conclude, citizens need to be enabled to discern and competently navigate public and commercial choice environments in which choice architects do not act benevolently and financial conflicts of interest are endemic. Without boosting people's skills to see through toxic choice environments, this will be barely possible, or its effects limited. Next, I discuss two somewhat neglected dimensions that should be entered into a cost–benefit analysis in order to determine which intervention maximises welfare.

Redesigning school cafeterias to promote healthier food choices is a paradigmatic example of a public policy nudge (Thaler & Sunstein, Reference Thaler and Sunstein2008), but school cafeterias are just one context in which high-school students make food choices. Each context has its own choice architecture, and many are not under the control of a benevolent choice architect. For instance, many students have countless opportunities to buy and consume junk food on their way home from school; once home, they are bombarded with television food advertisements.

If the goal is to promote healthy food choices that promise to be generalisable across a wide range of (benevolently or commercially constructed) choice architectures, including architectures that are harder to reach by nudges than by boosts (e.g. the family dinner table; Dallacker et al., Reference Dallacker, Hertwig and Mata2017), then boosts (e.g. in the form of a set of simple, memorable rules for eating wisely; Pollan, Reference Pollan2009) are more likely to succeed than nudges. Indeed, one of the advantages of boosts is that people can apply them independently of the given choice architecture. A nudge in the form of cafeteria design, in contrast, is restricted to the specific context in which it is implemented.

In theory, boosts aim to produce long-lasting behavioural effects by instilling a new competence or fostering an existing one. This is less clear in the case of nudges. One-time nudges (e.g. the organ donation opt-out default or automatic enrolment in a savings plan or green energy) may cease to have the desired behavioural effect as soon as they are removed. In contrast, many-time nudges that affect behaviour repeatedly (e.g. daily food choices in a rearranged cafeteria) may produce behavioural routines through learning (e.g. healthy food choices). These routines may ‘survive’ the removal of the initial choice architecture and generalise to other dietary decisions.

In the service of lasting behaviours, let me also emphasise the importance of self-nudging. Self-nudging means that people intentionally nudge themselves in order to self-regulate their behaviour and break self-destructive habits (akin to Thomas Schelling's self-commitment devices). For example, people can rearrange their kitchen so that sweets and snacks are placed out of convenient reach and healthier foods are stored at eye level. A user can set his or her browser to open with a news page rather than a sports page. When the nudger and the nudged are one and the same person, as in the case of self-nudging, autonomy and agency remain intact.

To nudge him or herself, however, the individual has to have the same insights into the nudging intervention and the underlying mechanism as the choice architect (unlike in the conception of nudging of Thaler & Sunstein, Reference Thaler and Sunstein2008). The nudge thus becomes a boost insofar as individuals first need to be educated about the rationale of the self-nudging intervention, its mechanisms and effects and how those effects can be harnessed. Indeed, this is the approach that is often taken in behaviour change programmes in health psychology (e.g. Michie et al., Reference Michie, Johnston, Francis, Hardeman and Eccles2008).

The risk of unintended and undesirable behavioural consequences (‘behavioural spillovers’; Dolan & Galizzi, Reference Dolan and Galizzi2015) is an insufficiently explored dimension of some forms of nudging. For example, past good deeds can liberate individuals to engage in behaviours that are immoral, unethical or otherwise problematic – behaviours that they would otherwise avoid. Something akin to this ‘moral self-licensing’ (e.g. Monin & Miller, Reference Monin and Miller2001) may also occur when people are nudged in certain ways. For instance, if a choice architecture nudges individuals to make healthier food choices by, for instance, rearranging foods in the cafeteria (order, eye level, etc.), people who had salad for lunch may feel liberated to consume more calories later (afternoon snacks or dinner) than they otherwise would have done. In the financial domain, automatic enrolment in pension plans has been shown to substantially increase participation (Chetty et al., Reference Chetty, Friedman, Leth-Petersen, Nielsen and Olsen2014), but people who feel safe in the knowledge that they are investing in their retirement may feel liberated to engage in more discretionary spending then they otherwise would have done.

Whether a nudge has such unintended negative behavioural effects is an empirical question, as is the magnitude of any such effects. Should they exist for a specific nudge, however, then a boost may be the more advisable intervention. Although it may be tempting to argue that boosts are less likely to have unintended negative consequences than nudges – because boosts are less likely to be designed as local repairs with unanticipated consequences in some other domain – the very nature of unintended consequences is that they are difficult to anticipate. Therefore, boosts, like nudges, need to be evaluated in terms of their intended and unintended consequences.

Finally, some unintended consequences may, of course, be desirable in nature. For instance, people who are nudged repeatedly to adopt healthier food choices may extend this behaviour to other domains (e.g. exercising). As with negative spillover effects, whether such effects occur – and for what forms of nudging or boosting – is an empirical question. If positive spillover effects occur, they would, of course, add a desirable and little-explored dimension to both forms of policy intervention.